Respiratory diseases, both acute and chronic, are widespread due to exposure to harmful substances in the environment, workplace, and through personal behaviors. Furthermore, the COVID-19 pandemic has led to both short-term and long-term lung damage in survivors. Therefore, accurate identification of chronic respiratory diseases, in particular, is vital for effective management and treatment. Auscultation, the practice of listening to respiratory sounds, plays a crucial role in diagnosing respiratory diseases. By accurately interpreting these sounds, complemented by other clinical findings, specialists can make reliable diagnoses with minimal errors. However, the effectiveness of auscultation is heavily influenced by the doctor's experience and environmental noise. To address these limitations, automatic classification of respiratory sounds recorded with a digital stethoscope using expert software has emerged as a popular research area. This approach eliminates the reliance on subjective interpretation by specialists. Unfortunately, as with many biomedical signals, researchers face significant challenges. The most pressing issue is the need for high-quality, accurately labeled, and extensive lung and respiratory sound datasets. Additionally, removing noise that distorts these sound signals is another major obstacle. This brief review aims to delve into these two primary challenges and provide examples of potential solutions from relevant literature.